Run NMT Translation#

Use this guide when translation should run against backend: nmt and your own HTTP microservice that performs neural machine translation (NMT).

Prerequisites#

  • A running NMT service reachable at the URL you pass as nmt.server_url, implementing the contract in the next section.

  • If faith_eval.enabled is true, LLM credentials for FAITH remain required; see Run FAITH Evaluation.

Service Contract#

Your service must implement one endpoint.

POST /translate — request body:

{
  "texts": ["Hello, world.", "How are you?"],
  "src_lang": "en",
  "tgt_lang": "hi"
}

texts is a list of strings (one per segment). src_lang and tgt_lang are BCP 47 / ISO 639-1 codes; Nemotron lowercases them before the call.

Success response body:

{
  "translations": ["नमस्ते, दुनिया।", "आप कैसे हैं?"]
}

translations must contain exactly as many strings as texts. Curator raises a RuntimeError on a count mismatch.

On error, return a non-2xx HTTP status with a JSON body containing an "error" field. Curator retries with exponential backoff and surfaces unrecoverable failures as an aiohttp.ClientResponseError. Streaming is not supported — return the full batch in a single response.

Curator also sends GET /health at startup as an optional liveness check. Return any 2xx to pass; if the endpoint is absent, Curator logs a warning and continues.

Tune nmt.batch_size, nmt.timeout, and nmt.max_concurrent_requests once you understand your server’s throughput.

Procedure#

uv run nemotron steps run translate/nemo_curator -c default \
  backend=nmt \
  nmt.server_url=http://localhost:5000 \
  input_path=/path/to/chat.jsonl \
  output_dir=/path/to/out \
  source_language=en \
  target_language=hi

FAITH Considerations#

If faith_eval.enabled stays true as in the default default.yaml, you must still supply large language model (LLM) credentials through server because FAITH scoring uses that client even though translation runs on NMT.